TypeSafe's Jev 1.14.0 unlocks multimodality, a new box type, and 262K context behind danger-zone flags
op7418 · x · 2026-10-15
TypeSafe AI shipped Jev 1.14.0 with three experimental "Danger Zone" capabilities enabled via API headers:
- Images: jev-1.13.0 was multimodal all along but support wasn't exposed. Images can now be composed anywhere in the JSON interface (state, instructions, criteria) via find-and-replace in assets. TypeSafe warns multimodal still underperforms plain text.
- box type: The first non-judgment output type — given an image and a query, it returns a guaranteed-valid bounding box for the target, only on jev-1.14.0. TypeSafe frames it as a preview of System One models moving beyond simple judgments.
- 262K context: default limits of 64k per request / 32k per question can be raised to 262k each with a header; the company says long context was deliberately capped as less reliable, recommending proper state management.
Multilingual quality also improved markedly. jev-1.14.0 ships as preview: jev-latest switches in 1 week, with 60/30-day deprecation notice periods for old versions.
More from Models
- 9 AI Models Summarize Edited Huckleberry Finn; Only Claude Followed Both Times — GaryMarcus · 2026-10-15
- Alexandr Wang hails new Muse model's writing and orchestration, naming and pricing panned — alexandr_wang · 2026-10-15
- Dev seeks free H200 to scale Luganda chat model Pearl Chat from 8.5M to 500M parameters — weskambale · 2026-10-15
- Blogger hopes for new open-weights Gemma or Meta models, citing open-source writing gap — ivan_bezdomny · 2026-10-15
- Claude drops six compliments in one conversation, user calls out sycophancy — philnash · 2026-10-15
- Gemini Argon outscored Meta's Muse Wonder in news-writing tests, but both have their own voice — ivan_bezdomny · 2026-10-15